docs(log 038): the results — two masters, 14 tracks, and a clean negative on the global pass

Everything landed. Two full masters differing only in air shelf, both at -14.0 LUFS
/ -0.9 dBTP / LRA 7.5-7.6 / 44100 Hz 24-bit, and 14 split tracks verifying ALL OK.
Duration is 3779.317551 s against a source of 3779.317550 — the load-bearing number,
because it means every one of PLN's playhead calls stays valid and the split lands
exactly where he put the cuts rather than approximately there.

Q1: the missing air is CYMBALS. Drums hold 0.367% of their own energy above 8 kHz
against 0.084% for the runner-up, a 4.4x concentration where evenly-spread energy
would have meant codec hiss. So opening the shelf is licensed by measurement, and
how far is not — hence an A/B rather than a number I liked. The framing that
matters: both candidates are still BELOW typical brightness for electronic masters
(8-16 kHz at 0.40% and 0.63% against a 1-3% norm), so the choice is "quite dark vs
slightly less dark", not "safe vs aggressive". The A/B is also confound-free:
identical loudness, identical true peak, range within 0.1 LU, only the tilt differs.

Q2: the floor is worse here than OPAL. Kick active time averages 44.8% against the
59% project_floor_problem measured for OPAL-26 from real orbit stems, worst at
#2 22.8%, #3 33.4%, #8 36.9%. A stage mic in a room explains it on its own, so this
is a lead rather than a verdict — but it is the first time this gig's floor has been
measurable at all.

Q3, PLN's own question, and the answer is a clean negative: the global pass was not
worth running. Correlations 0.943-0.977 against the per-section stems with level
deltas of at most 0.21 dB. The two separations are the same stems. The prediction
held exactly — htdemucs' 7.8 s receptive field means a global pass cannot give the
model more musical context, so only normalisation differs, and it barely moved. It
cost ~2.5 h of CPU competing with the master render to establish a quarter of a dB.
Sections only next time; the question was worth asking once, not twice.

Side finding: bass is the least stably separated of the four sources (0.9432,
lowest), consistent with the same run putting PunkAChien's bass into `other` — bass
stem at -81.0 dBFS while `other` was the loudest stem in the set.

The log also records three self-inflicted errors that share one shape: checks whose
failure mode was indistinguishable from success. A dropped rtk grep plus a
block-buffered logfile that made a working job look dead, so two renders raced on
the same paths for forty minutes. A speed claim built on counting directories that
appear when writers open. And a scripted edit that silently no-op'd while printing
success unconditionally. Plus the 192 kHz master, which was not a new bug at all —
topic_postprod_mastering has carried that trap since tidal-ears hit it, and I wrote
a new mastering chain without re-reading the omnibus that exists to prevent it.

armada/tasks/038-a-record-from-one-microphone.md

@@ -274,3 +274,102 @@ Branch `claude/release-board`.
 - Whether `Outro: La Dub Sirène` (103 s) is an album track or mix-only.
 - Whether the club target is worth chasing below the LRA floor on this source.
 - The unreviewed merge `2f947c2`, still standing on master.
+
+---
+
+## The results — 2026-09-02, ~02:00
+
+### Delivered
+
+| | |
+|---|---|
+| **Master A** `Cosmic26_v1_streaming.flac` | air shelf **+5 dB** · 661 MB |
+| **Master B** `ab_air8/streaming.flac` | air shelf **+8 dB** · 673 MB |
+| both | **−14.0 LUFS · −0.9 dBTP · LRA 7.5 / 7.6 · 44100 Hz / 24-bit** |
+| duration | **3779.317551 s** against a source of 3779.317550 s |
+| **album split** | `tracks_v1_streaming/` — 14 FLACs, verify **ALL OK**, format 44100/24 |
+| demucs | 14 sections + a full 3779 s global pass |
+
+Both of PLN's forms are therefore on disk: nothing was cut from this set, so the
+continuous mix simply *is* the master file, and the split sits beside it.
+
+The duration matching the source to within a microsecond is the load-bearing
+number. It means every one of his playhead calls stays valid against the master,
+so the split lands exactly where he put the cuts rather than approximately there.
+
+### Q1 — the missing air is CYMBALS
+
+Mean share of each stem's **own** energy above 8 kHz: drums **0.367%**, vocals
+0.084%, other 0.033%, bass 0.000%. A **4.4× concentration in drums**, where evenly
+spread energy would have meant codec hiss and a shelf that must stay shut.
+
+So the deficit is real drum content the room ate, and opening the shelf is licensed
+by measurement. How far is not, so the answer is an A/B rather than a number I
+liked. The useful context for that choice: against the tape, 4–8 kHz went 1.75% →
+3.14% (+5) → 3.86% (+8), and 8–16 kHz went 0.12% → 0.40% → 0.63%, against a rough
+norm of 3–6% and 1–3% for electronic masters. **Both candidates are still below
+typical for air.** The choice is "quite dark vs slightly less dark", not "safe vs
+aggressive" — worth saying, because the framing changes the answer.
+
+The A/B is also clean in the way that matters: identical loudness (−14.0), identical
+true peak (−0.9), range within 0.1 LU. Only the tilt differs — 0.98× below 1 kHz,
+1.23× at 4–8 k, 1.59× at 8–16 k. No "the louder one sounds better" confound.
+
+### Q2 — the floor is worse here than OPAL
+
+Kick active time, 40–120 Hz against each track's own p95: **mean 44.8%**, against
+the **59%** `project_floor_problem` measured for OPAL-26 *from real orbit stems*.
+Worst: **#2 There's Something About Drums 22.8%**, **#3 Am i Doing it Right 33.4%**,
+**#8 Eh ouais je Funk 36.9%**. A stage mic in a room is a sufficient explanation on
+its own, so this is a lead and not a verdict — but it is the first time this gig's
+floor has been measurable at all.
+
+### Q3 — the global pass was NOT worth running
+
+PLN's own question, and the answer is a clean negative:
+
+| stem | correlation | level delta |
+|---|---|---|
+| drums | 0.9765 | −0.14 dB |
+| other | 0.9742 | −0.06 dB |
+| vocals | 0.9543 | −0.17 dB |
+| bass | **0.9432** | −0.21 dB |
+
+**The two separations are the same stems.** The prediction held exactly: htdemucs'
+7.8 s receptive field means a global pass cannot give the model more musical
+context, so only the normalisation statistics differ — and they moved the level by
+at most 0.21 dB. The residual 2–6% of uncorrelated energy is chunk-boundary
+placement, not a different reading of the music.
+
+It cost roughly 2.5 hours of CPU, competing with the master render, to establish a
+≤0.21 dB difference. **Sections only on the next stemless gig.** The question was
+worth asking once; it is not worth paying for twice. Banked in
+`reference_demucs_on_xps22`.
+
+Side finding: **bass is the least stably separated source** (0.9432, lowest of the
+four), which is consistent with the same run putting PunkAChien's bass into
+`other` — bass stem at −81.0 dBFS while `other` was the loudest stem in the set.
+Anyone reading stem filenames as roles would have concluded that track has no bass.
+
+### Three self-inflicted errors, same shape
+
+All three were **checks whose failure mode was indistinguishable from success**:
+
+1. **The nohup "death".** `rtk` silently dropped a `grep`, and Python block-buffered
+   a logfile to zero bytes. I concluded the job was dead and started a second copy;
+   two renders then wrote the same paths for ~40 minutes.
+2. **The demucs speed claim.** Counted output directories, which appear when writers
+   *open*. Retracted from the commit, the log, and two memory files.
+3. **A silent no-op edit.** A scripted replacement whose search string carried
+   escaped line-continuations that did not match the file, while the script printed
+   "gate tightened" unconditionally. Caught by grepping the file rather than
+   trusting the message.
+
+And one that was not mine to invent but was mine to prevent: the first full master
+came out at **192 kHz, 1.32 GB** because ffmpeg's `loudnorm` oversamples for
+true-peak detection and never resamples back. `topic_postprod_mastering` has carried
+that exact trap since tidal-ears hit it. I wrote a new mastering chain without
+re-reading the omnibus that exists to prevent it. The fix is structural — the
+resample lives in the chain *builder*, and the in-spec gate now asserts sample rate
+and channel count beside I/TP/LRA, because a loudness-only gate certified a
+4.3×-oversized master as ✓.